Automating Resilience Tuning

Abstract

Resilience frameworks are often used in microservice software ecosystems to ensure that failure in a service does not propagate throughout the rest of the ecosystem. These frameworks need to be tuned for each connection they are applied to, which is a time-consuming task. In this work, the possibilities of using discrete event simulation to help automate the tuning process of resilience frameworks in microservice software ecosystems is studied. A simulation model is designed, based on the microservice ecosystem at bol.com, a large online retailer in the Netherlands. The model simulates a connection between two services in the bol.com ecosystem. The model is improved upon iteratively during the project as a result of new insights gained from interviews with domain experts at bol.com and results of experiments. Using system logs available at bol.com, an input analysis is performed. In collaboration with domain experts at bol.com, performance measures are designed. Finally, this simulation model is then used by an optimization heuristic which tests different configurations. The results of the experiments show some interesting traits of the configuration parameters and relations they may or may not have with the effectiveness of the resilience framework.